Structure Optimization of Radial Basis Probabilistic Neural Networks by the Maximum Absolute Error Combined with the Micro-Genetic Algorithm

The maximum absolute error algorithm (MAEA) is used to optimally selecting the hidden centers vectors of the radial basis probabilistic neural networks (RBPNN). The MAEA is combined with the micro-genetic algorithm (μGA), which is used to optimize the controlling parameter of the kernel function of the RBPNN, i.e., MAE-μGA, so as to carry out optimizing the overall structure of RBPNN. The experiments demonstrate that the RBPNN, optimized by the MAE-μGA, has the best simple structure compared with results by the other optimization methods introduced. Furthermore, in the aspect of the generalization performance of the optimized networks, the RBPNN by the MAE-μGA is a little better than ones by the other methods. In addition, the MAE-μGA can also be used to optimize the radial basis function neural networks (RBFNN).